Skip to contentSkip to conversation

BUG: soft lockup during suspend

DISCUSSION Pavel Machek 3 participants 11 messages Mar 20, 2007 – Mar 26, 2007 Sound

lore
Share & tools
lore
#0 lore raw

Hi!

I got this nastinness in my syslog... perhaps HDA intel takes too long to play with its hardware? Or should we just kill the softlockup watchdog since Linux is not realtime system, yet?
Pavel

HDA Intel 0000:00:1b.0: freeze

BUG: soft lockup detected on CPU#0!
 [<c014fdd9>] softlockup_tick+0xa9/0xd0
 [<c0131d63>] update_process_times+0x33/0x80
 [<c0142f52>] tick_periodic+0x22/0x70
 [<c0142fb7>] tick_handle_periodic+0x17/0x80
 [<c014359a>] tick_do_broadcast+0x6a/0x80
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01435fb>] tick_handle_periodic_broadcast+0x1b/0x60
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01077cc>] timer_interrupt+0x2c/0x40
 [<c0150005>] handle_IRQ_event+0x25/0x60
 [<c0151d87>] handle_edge_irq+0xe7/0x130
 [<c0106adb>] do_IRQ+0x3b/0x80
 [<c0106ae0>] do_IRQ+0x40/0x80
 [<c01049eb>] common_interrupt+0x23/0x28
 [<c063007b>] ieee80211_wx_get_scan+0x2b/0x130
 [<c02588f4>] delay_tsc+0x14/0x20
 [<c0258936>] __delay+0x6/0x10
 [<c052f3da>] azx_send_cmd+0xfa/0x110
 [<c053089e>] snd_hda_codec_write+0x3e/0x60
 [<c05314fb>] hda_set_power_state+0xab/0xe0
 [<c0532518>] snd_hda_suspend+0x48/0x60
 [<c0530302>] azx_suspend+0x52/0xd0
 [<c0266b13>] pci_device_suspend+0x23/0x70
 [<c0332dbb>] suspend_device+0x11b/0x2e0
 [<c0333032>] device_suspend+0xb2/0x170
 [<c012a28b>] printk+0x1b/0x20
 [<c014c7ad>] pm_suspend_disk+0x4d/0x2b0
 [<c014b395>] enter_state+0x195/0x220
 [<c014b4c9>] state_store+0xa9/0xc0
 [<c014b420>] state_store+0x0/0xc0
 [<c01ae789>] subsys_attr_store+0x29/0x40
 [<c01af07c>] sysfs_write_file+0xac/0x130
 [<c0172156>] vfs_write+0xa6/0x140
 [<c01aefd0>] sysfs_write_file+0x0/0x130
 [<c0104050>] syscall_call+0x7/0xb
 [<c0630000>] ieee80211_translate_scan+0xa00/0xa50
 =======================

ACPI: PCI interrupt for device 0000:00:1b.0 disabled

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#1 lore raw

At Tue, 20 Mar 2007 13:32:53 +0100,

Quoted 6 lines · Pavel Machek

Hi!

I got this nastinness in my syslog... perhaps HDA intel takes too long to play with its hardware? Or should we just kill the softlockup watchdog since Linux is not realtime system, yet?

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

Takashi

Quoted 47 lines

Pavel

HDA Intel 0000:00:1b.0: freeze

BUG: soft lockup detected on CPU#0!
 [<c014fdd9>] softlockup_tick+0xa9/0xd0
 [<c0131d63>] update_process_times+0x33/0x80
 [<c0142f52>] tick_periodic+0x22/0x70
 [<c0142fb7>] tick_handle_periodic+0x17/0x80
 [<c014359a>] tick_do_broadcast+0x6a/0x80
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01435fb>] tick_handle_periodic_broadcast+0x1b/0x60
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01077cc>] timer_interrupt+0x2c/0x40
 [<c0150005>] handle_IRQ_event+0x25/0x60
 [<c0151d87>] handle_edge_irq+0xe7/0x130
 [<c0106adb>] do_IRQ+0x3b/0x80
 [<c0106ae0>] do_IRQ+0x40/0x80
 [<c01049eb>] common_interrupt+0x23/0x28
 [<c063007b>] ieee80211_wx_get_scan+0x2b/0x130
 [<c02588f4>] delay_tsc+0x14/0x20
 [<c0258936>] __delay+0x6/0x10
 [<c052f3da>] azx_send_cmd+0xfa/0x110
 [<c053089e>] snd_hda_codec_write+0x3e/0x60
 [<c05314fb>] hda_set_power_state+0xab/0xe0
 [<c0532518>] snd_hda_suspend+0x48/0x60
 [<c0530302>] azx_suspend+0x52/0xd0
 [<c0266b13>] pci_device_suspend+0x23/0x70
 [<c0332dbb>] suspend_device+0x11b/0x2e0
 [<c0333032>] device_suspend+0xb2/0x170
 [<c012a28b>] printk+0x1b/0x20
 [<c014c7ad>] pm_suspend_disk+0x4d/0x2b0
 [<c014b395>] enter_state+0x195/0x220
 [<c014b4c9>] state_store+0xa9/0xc0
 [<c014b420>] state_store+0x0/0xc0
 [<c01ae789>] subsys_attr_store+0x29/0x40
 [<c01af07c>] sysfs_write_file+0xac/0x130
 [<c0172156>] vfs_write+0xa6/0x140
 [<c01aefd0>] sysfs_write_file+0x0/0x130
 [<c0104050>] syscall_call+0x7/0xb
 [<c0630000>] ieee80211_translate_scan+0xa00/0xa50
 =======================

ACPI: PCI interrupt for device 0000:00:1b.0 disabled

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#2 lore raw

Hi!

Quoted 10 lines
Quoted 3 lines

I got this nastinness in my syslog... perhaps HDA intel takes too long to play with its hardware? Or should we just kill the softlockup watchdog since Linux is not realtime system, yet?

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

Is adding touch_softlockup_watchdog() to hd_audio right solution? Or should watchdog just disappear?
Pavel

Quoted 45 lines
Quoted 45 lines

HDA Intel 0000:00:1b.0: freeze

BUG: soft lockup detected on CPU#0!
 [<c014fdd9>] softlockup_tick+0xa9/0xd0
 [<c0131d63>] update_process_times+0x33/0x80
 [<c0142f52>] tick_periodic+0x22/0x70
 [<c0142fb7>] tick_handle_periodic+0x17/0x80
 [<c014359a>] tick_do_broadcast+0x6a/0x80
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01435fb>] tick_handle_periodic_broadcast+0x1b/0x60
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01077cc>] timer_interrupt+0x2c/0x40
 [<c0150005>] handle_IRQ_event+0x25/0x60
 [<c0151d87>] handle_edge_irq+0xe7/0x130
 [<c0106adb>] do_IRQ+0x3b/0x80
 [<c0106ae0>] do_IRQ+0x40/0x80
 [<c01049eb>] common_interrupt+0x23/0x28
 [<c063007b>] ieee80211_wx_get_scan+0x2b/0x130
 [<c02588f4>] delay_tsc+0x14/0x20
 [<c0258936>] __delay+0x6/0x10
 [<c052f3da>] azx_send_cmd+0xfa/0x110
 [<c053089e>] snd_hda_codec_write+0x3e/0x60
 [<c05314fb>] hda_set_power_state+0xab/0xe0
 [<c0532518>] snd_hda_suspend+0x48/0x60
 [<c0530302>] azx_suspend+0x52/0xd0
 [<c0266b13>] pci_device_suspend+0x23/0x70
 [<c0332dbb>] suspend_device+0x11b/0x2e0
 [<c0333032>] device_suspend+0xb2/0x170
 [<c012a28b>] printk+0x1b/0x20
 [<c014c7ad>] pm_suspend_disk+0x4d/0x2b0
 [<c014b395>] enter_state+0x195/0x220
 [<c014b4c9>] state_store+0xa9/0xc0
 [<c014b420>] state_store+0x0/0xc0
 [<c01ae789>] subsys_attr_store+0x29/0x40
 [<c01af07c>] sysfs_write_file+0xac/0x130
 [<c0172156>] vfs_write+0xa6/0x140
 [<c01aefd0>] sysfs_write_file+0x0/0x130
 [<c0104050>] syscall_call+0x7/0xb
 [<c0630000>] ieee80211_translate_scan+0xa00/0xa50
 =======================

ACPI: PCI interrupt for device 0000:00:1b.0 disabled

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#3 lore raw

At Tue, 20 Mar 2007 14:22:03 +0100,

Quoted 14 lines · Pavel Machek

Hi!

Quoted 10 lines
Quoted 3 lines

I got this nastinness in my syslog... perhaps HDA intel takes too long to play with its hardware? Or should we just kill the softlockup watchdog since Linux is not realtime system, yet?

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

Quoted 2 lines

Is adding touch_softlockup_watchdog() to hd_audio right solution? Or should watchdog just disappear?

This should be at a loop in azx_single_send_cmd(),

	int timeout = 50;

	...
	while (timeout--) {
		/* check ICB busy bit */
		if (! (azx_readw(chip, IRS) & ICH6_IRS_BUSY)) {
			...
			return 0;
		}
		udelay(1);
	}

and this function is not in spinlock by itself.
Hence I feel the softlockup is too sensitive in this regard. But calling touch_softlockup_watchdog() is surely a workaround.

Takashi

Quoted 52 lines

Pavel

Quoted 45 lines
Quoted 45 lines

HDA Intel 0000:00:1b.0: freeze

BUG: soft lockup detected on CPU#0!
 [<c014fdd9>] softlockup_tick+0xa9/0xd0
 [<c0131d63>] update_process_times+0x33/0x80
 [<c0142f52>] tick_periodic+0x22/0x70
 [<c0142fb7>] tick_handle_periodic+0x17/0x80
 [<c014359a>] tick_do_broadcast+0x6a/0x80
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01435fb>] tick_handle_periodic_broadcast+0x1b/0x60
 [<c01435cc>] tick_do_periodic_broadcast+0x1c/0x30
 [<c01077cc>] timer_interrupt+0x2c/0x40
 [<c0150005>] handle_IRQ_event+0x25/0x60
 [<c0151d87>] handle_edge_irq+0xe7/0x130
 [<c0106adb>] do_IRQ+0x3b/0x80
 [<c0106ae0>] do_IRQ+0x40/0x80
 [<c01049eb>] common_interrupt+0x23/0x28
 [<c063007b>] ieee80211_wx_get_scan+0x2b/0x130
 [<c02588f4>] delay_tsc+0x14/0x20
 [<c0258936>] __delay+0x6/0x10
 [<c052f3da>] azx_send_cmd+0xfa/0x110
 [<c053089e>] snd_hda_codec_write+0x3e/0x60
 [<c05314fb>] hda_set_power_state+0xab/0xe0
 [<c0532518>] snd_hda_suspend+0x48/0x60
 [<c0530302>] azx_suspend+0x52/0xd0
 [<c0266b13>] pci_device_suspend+0x23/0x70
 [<c0332dbb>] suspend_device+0x11b/0x2e0
 [<c0333032>] device_suspend+0xb2/0x170
 [<c012a28b>] printk+0x1b/0x20
 [<c014c7ad>] pm_suspend_disk+0x4d/0x2b0
 [<c014b395>] enter_state+0x195/0x220
 [<c014b4c9>] state_store+0xa9/0xc0
 [<c014b420>] state_store+0x0/0xc0
 [<c01ae789>] subsys_attr_store+0x29/0x40
 [<c01af07c>] sysfs_write_file+0xac/0x130
 [<c0172156>] vfs_write+0xa6/0x140
 [<c01aefd0>] sysfs_write_file+0x0/0x130
 [<c0104050>] syscall_call+0x7/0xb
 [<c0630000>] ieee80211_translate_scan+0xa00/0xa50
 =======================

ACPI: PCI interrupt for device 0000:00:1b.0 disabled

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#4 lore raw
In reply to Takashi Iwai
Quoted 7 lines · Takashi Iwai

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

What about Acer Aspire 5102?

http://lkml.org/lkml/2006/12/3/9

#5 lore raw

At Tue, 20 Mar 2007 12:05:07 -0400,

Quoted 10 lines · Chuck Ebbert
Quoted 7 lines · Takashi Iwai

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

Quoted 3 lines

What about Acer Aspire 5102?

http://lkml.org/lkml/2006/12/3/9

This is a different problem.
A known workaround is to provide probe_mask=1 module option.

Takashi

#6 lore raw

At Tue, 20 Mar 2007 17:08:48 +0100,

Quoted 13 lines · I wrote:

At Tue, 20 Mar 2007 12:05:07 -0400,

Quoted 10 lines · Chuck Ebbert
Quoted 7 lines · Takashi Iwai

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

Quoted 6 lines
Quoted 3 lines

What about Acer Aspire 5102?

http://lkml.org/lkml/2006/12/3/9

This is a different problem.
A known workaround is to provide probe_mask=1 module option.

BTW, does this happen on the latest linus git tree? (rc4 may work, though)

If yes, could you try the patch below?

thanks,

Takashi

sound/pci/hda/hda_intel.c +14 -2
diff -r 1dd843b6ffa7 sound/pci/hda/hda_intel.c
@@ -198,6 +198,7 @@ enum { SDI0, SDI1, SDI2, SDI3, SDO0, SDO
 #define RIRB_INT_MASK		0x05
 
 /* STATESTS int mask: SD2,SD1,SD0 */
+#define AZX_MAX_CODECS		3
 #define STATESTS_INT_MASK	0x07
 
 /* SD_CTL bits */
@@ -991,7 +992,7 @@ static int __devinit azx_codec_create(st
 		return err;
 
 	codecs = 0;
-	for (c = 0; c < azx_max_codecs[chip->driver_type]; c++) {
+	for (c = 0; c < AZX_MAX_CODECS; c++) {
 		if ((chip->codec_mask & (1 << c)) & probe_mask) {
 			err = snd_hda_codec_new(chip->bus, c, NULL);
 			if (err < 0)
@@ -999,7 +1000,18 @@ static int __devinit azx_codec_create(st
 			codecs++;
 		}
 	}
-	if (! codecs) {
+	if (!codecs) {
+		/* probe additional slots if no codec is found */
+		for (; c < azx_max_codecs[chip->driver_type]; c++) {
+			if ((chip->codec_mask & (1 << c)) & probe_mask) {
+				err = snd_hda_codec_new(chip->bus, c, NULL);
+				if (err < 0)
+					continue;
+				codecs++;
+			}
+		}
+	}
+	if (!codecs) {
 		snd_printk(KERN_ERR SFX "no codecs initialized\n");
 		return -ENXIO;
 	}
#7 lore raw
Quoted 13 lines · Takashi Iwai

At Tue, 20 Mar 2007 17:08:48 +0100,

Quoted 11 lines · I wrote:

At Tue, 20 Mar 2007 12:05:07 -0400,

Quoted 9 lines · Chuck Ebbert
Quoted 6 lines · Takashi Iwai

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

What about Acer Aspire 5102?

Quoted 9 lines
Quoted 3 lines
Quoted 1 line

http://lkml.org/lkml/2006/12/3/9

This is a different problem.
A known workaround is to provide probe_mask=1 module option.

BTW, does this happen on the latest linus git tree? (rc4 may work, though)

If yes, could you try the patch below?

I can't easily test 2.6.21-rc, but "probe_mask=1" works on 2.6.20.4, which spews errors without that. Should I try your patch on 2.6.20?

#8 lore raw

At Tue, 20 Mar 2007 12:32:50 -0400,

Quoted 19 lines · Chuck Ebbert
Quoted 17 lines · Takashi Iwai

At Tue, 20 Mar 2007 17:08:48 +0100,

Quoted 14 lines · I wrote:

At Tue, 20 Mar 2007 12:05:07 -0400,

Quoted 10 lines · Chuck Ebbert
Quoted 6 lines · Takashi Iwai

X60/T60 is known to be often broken regarding the communication between the controller and the codec chip. When this kind of thing happens, the driver tries to switch to a single-shot I/O without using ring-buffers and IRQs, and even in such a mode, the communication gets broken. FWIW, it doesn't happen on other machines with HD-audio, so it's fairly specific to X60/T60. No idea why.

What about Acer Aspire 5102?

http://lkml.org/lkml/2006/12/3/9

This is a different problem.
A known workaround is to provide probe_mask=1 module option.

Quoted 8 lines
Quoted 5 lines

BTW, does this happen on the latest linus git tree? (rc4 may work, though)

If yes, could you try the patch below?

I can't easily test 2.6.21-rc, but "probe_mask=1" works on 2.6.20.4, which spews errors without that. Should I try your patch on 2.6.20?

Well, 2.6.21 has really tons of fixes/improvements regarding HD-audio, so I basically prefer tests on 2.6.21. But, at least, the patch can be applied on the top of Greg's latest 2.6.20.x patches...

Takashi

#9 lore raw
In reply to Takashi Iwai
Quoted 11 lines · Takashi Iwai

At Tue, 20 Mar 2007 17:08:48 +0100,

Quoted 5 lines · I wrote:

At Tue, 20 Mar 2007 12:05:07 -0400,

Quoted 2 lines

http://lkml.org/lkml/2006/12/3/9

This is a different problem.
A known workaround is to provide probe_mask=1 module option.

BTW, does this happen on the latest linus git tree? (rc4 may work, though)

2.6.21-rc4 works perfectly, detecting my Acer notebook's model properly and adding the sound devices to the bus. Very nice...

#10 lore raw

At Fri, 23 Mar 2007 15:22:46 -0400,

Quoted 10 lines · Chuck Ebbert
Quoted 8 lines · Takashi Iwai

At Tue, 20 Mar 2007 17:08:48 +0100,

Quoted 5 lines · I wrote:

At Tue, 20 Mar 2007 12:05:07 -0400,

Quoted 2 lines

http://lkml.org/lkml/2006/12/3/9

This is a different problem.
A known workaround is to provide probe_mask=1 module option.

Quoted 6 lines
Quoted 3 lines

BTW, does this happen on the latest linus git tree? (rc4 may work, though)

2.6.21-rc4 works perfectly, detecting my Acer notebook's model properly and adding the sound devices to the bus. Very nice...

Could you try rc5? I'm afraid that one of fix patches on rc5 may break the probing on your laptop (this path had to be applied to recover another regression...)

If it happened, try the patch below from the latest ALSA tree.

thanks,

Takashi

[PATCH] [ALSA] hda-intel - Probe additional slots only if necessary

Probing the codec slots on ATI controller causes problems on some devices like Acer laptops. On these devices, reading from codec slot 3 results in the communication failure with the codec chip. Meanwhile, some laptops (e.g. Gateway) have the codec connection only on slot 3, and probing this slot is mandatory for them. The patch improves the probing robustness. The additional slots are now checked only when no codecs are found in the primary three slots.

---
commit a239a9ff89307a0a65fd9cbb02afb020bdb13de0
tree 72603176c199d3fedaeba030f3ea5691c4b5d882
parent 66753e652a094e0ab856417a5092297dbabe5fd6
author Takashi Iwai <[email protected]> Wed, 21 Mar 2007 15:14:35 +0100 committer Jaroslav Kysela <[email protected]> Wed, 21 Mar 2007 15:15:24 +0100

 sound/pci/hda/hda_intel.c |   16 ++++++++++++++--
 1 files changed, 14 insertions(+), 2 deletions(-)
sound/pci/hda/hda_intel.c +14 -2
index 517a8d7..5e478b9 100644
@@ -198,6 +198,7 @@ enum { SDI0, SDI1, SDI2, SDI3, SDO0, SDO
 #define RIRB_INT_MASK		0x05
 
 /* STATESTS int mask: SD2,SD1,SD0 */
+#define AZX_MAX_CODECS		3
 #define STATESTS_INT_MASK	0x07
 
 /* SD_CTL bits */
@@ -991,7 +992,7 @@ static int __devinit azx_codec_create(st
 		return err;
 
 	codecs = 0;
-	for (c = 0; c < azx_max_codecs[chip->driver_type]; c++) {
+	for (c = 0; c < AZX_MAX_CODECS; c++) {
 		if ((chip->codec_mask & (1 << c)) & probe_mask) {
 			err = snd_hda_codec_new(chip->bus, c, NULL);
 			if (err < 0)
@@ -999,7 +1000,18 @@ static int __devinit azx_codec_create(st
 			codecs++;
 		}
 	}
-	if (! codecs) {
+	if (!codecs) {
+		/* probe additional slots if no codec is found */
+		for (; c < azx_max_codecs[chip->driver_type]; c++) {
+			if ((chip->codec_mask & (1 << c)) & probe_mask) {
+				err = snd_hda_codec_new(chip->bus, c, NULL);
+				if (err < 0)
+					continue;
+				codecs++;
+			}
+		}
+	}
+	if (!codecs) {
 		snd_printk(KERN_ERR SFX "no codecs initialized\n");
 		return -ENXIO;
 	}